ON THE APPLICATION AND COMPARISON OF CONDITION-BASED MAINTENANCE PROGNOSTICS APPROACHES: LOGICAL ANALYSIS OF DATA, ARTIFICIAL NEURAL NETWORKS, AND PROPORTIONAL HAZARDS MODELS
Abstract
Condition-based maintenance (CBM) is becoming increasingly important because it performs more efficient diagnoses and prognoses based on equipment condition compared to time-based methods. CBM models greatly inform maintenance decisions. This thesis examines three CBM fault prognostics models: artificial neural networks (ANNs), logical analysis of data (LAD), and Proportional Hazard Models (PHM). ANNs are layered models that imitate the human brain. LAD is a non-statistical pattern recognition technique. PHM statistically relates equipment age, condition, and hazard rate. A methodology is developed to apply and compare these models, and used on NASA’s Turbofan Engine Degradation Simulation Dataset and the structural health monitoring dataset from Halifax’s A. Murray MacKay Bridge. Results are evaluated using three metrics: error, half-life error, and a cost score. This thesis concludes that the LAD and feedforward ANN models compares favourably to the PHM. However, the feedback ANN has lower performance because of large variability in predictions.